U.S. patent application number 11/131867 was filed with the patent office on 2006-11-23 for device incorporating improved text input mechanism.
Invention is credited to Ramin O. Assadollahi.
Application Number | 20060265208 11/131867 |
Document ID | / |
Family ID | 36950052 |
Filed Date | 2006-11-23 |
United States Patent
Application |
20060265208 |
Kind Code |
A1 |
Assadollahi; Ramin O. |
November 23, 2006 |
Device incorporating improved text input mechanism
Abstract
A device including a display screen for displaying m-words of
data, a text entry device for entering data, a processor receiving
data from the text entry device and causing it to be displayed on
the display screen. Upon activation the processor initializes a
precursor to a predefined value. The device further includes a
non-volatile memory storing a dictionary containing a plurality of
entries, each entry including an index, a candidate word, and a
score. The processor selects a list of n-number of candidate words
from the dictionary whose index matches the precursor, and causes
m-number of candidate words from the list of candidate words to be
displayed on the display screen. The processor causes the display
to prompt the user to select one of the displayed candidate words
or enter a desired word using the text entry device.
Inventors: |
Assadollahi; Ramin O.;
(Konstanz, DE) |
Correspondence
Address: |
JONATHAN D. FEUCHTWANG
1330 CENTRAL AVENUE
DEERFIELD
IL
60015
US
|
Family ID: |
36950052 |
Appl. No.: |
11/131867 |
Filed: |
May 18, 2005 |
Current U.S.
Class: |
704/9 |
Current CPC
Class: |
G06F 40/242 20200101;
G06F 3/0237 20130101; G06F 40/274 20200101 |
Class at
Publication: |
704/009 |
International
Class: |
G06F 17/27 20060101
G06F017/27 |
Claims
1. A device, comprising: a display screen for displaying m-words of
data; a text entry device for entering data; a processor receiving
data from the text entry device and causing it to be displayed on
the display screen, wherein upon activation said processor
initializes a precursor to a predefined value; a non-volatile
memory storing a dictionary containing a plurality of entries, each
said entry including an index, a candidate word; said processor
selecting a list of n-number of candidate words from said
dictionary whose index matches the precursor; said processor
causing m-number of candidate words from said list of candidate
words to be displayed on said display screen; said processor
causing the display to prompt the user to select one of the
displayed candidate words or enter a desired word using said text
entry device.
2. The device according to claim 1, wherein in response to the
selection of a candidate word, said processor updates the precursor
by appending the selected candidate word to the end of the
precursor and deleting the leading word, selects a new list of
candidate words whose index matches the precursor and causes the
display to prompt the user to either select one of the displayed
candidate words or enter a desired word using said text entry
device.
3. The device according to claim 1, wherein said dictionary
containing a score associated with at least selected ones of the
plurality of entries, said candidate words are displayed on the
display screen in an order corresponding to their respective
score.
4. The device according to claim 1, wherein said dictionary
containing a score associated with at least selected ones of the
plurality of entries; and wherein if m, the number of words capable
of being displayed on the display screen is less than n, the number
of candidate words retrieved, then said processor causes m-number
of candidate words having the highest score from said list of
candidate words to be displayed on said display screen.
5. The device according to claim 1, wherein: said precursor
comprises at least one word; and if m, the number of words capable
of being displayed on the display screen is greater than n, the
number of candidate words retrieved, then said processor causes
n-number of candidate words from said list of candidate words to be
displayed on said display screen, and if said precursor is greater
than or equal 2 then the processor deletes the leading word from
the precursor, selects a supplemental list of candidate words from
said dictionary whose index matches the precursor, and displays m-n
number of said supplemental candidate words to be displayed on said
display screen.
6. The device according to claim 5, wherein said dictionary
containing a score associated with at least selected ones of the
plurality of entries; and wherein said candidate words are
displayed on the display screen in an order corresponding to their
respective score.
7. The device according to claim 1, wherein said dictionary
containing a score associated with at least selected ones of the
plurality of entries, and said processor periodically updates the
score of selected said candidate words stored in said non-volatile
memory to reflect a frequency of usage.
8. The device according to claim 1, wherein if said processor is
unable to find a match between at least one said index stored in
said dictionary and said precursor, then said processor creates a
new entry in said dictionary in which said precursor is the index
and the candidate word is the word entered using said text entry
device.
9. The device according to claim 1, wherein the processor causes
selected said candidate words from the selected list to be
displayed whose initial character(s) match the character(s) entered
using said text entry device.
10. The device according to claim 1, wherein said dictionary
containing a score associated with at least selected ones of the
plurality of entries; and wherein said processor causes the display
screen to display only candidate words whose score exceeds a
threshold value.
11. The device according to claim 1, wherein: said precursor
comprises at least one word; and if n, the number of candidate
words retrieved is a threshold number of words greater than m, the
number of words capable of being displayed on the display screen,
then said processor increases the size of the precursor by one
word, and selects a list of candidate words from said dictionary
whose index matches the precursor.
12. A method for inputting text to be displayed on a display screen
capable of displaying m-words of data, comprising the steps:
initializing a precursor to a predefined value; receiving text
input data and causing it to be displayed on the display screen;
providing a dictionary containing a plurality of entries, each said
entry including an index, a candidate word; selecting, from the
dictionary, a list of n-number of candidate words whose index
matches the precursor, the dictionary containing a plurality of
entries, each said entry including an index, a candidate word;
causing m-number of candidate words from said list of candidate
words to be displayed on said display screen; and prompting the
user to select one of the displayed candidate words or enter a
desired word.
13. The method according to claim 12, wherein said dictionary
containing a score associated with at least selected ones of the
plurality of entries; and wherein said candidate words are
displayed on the display screen in an order corresponding to their
respective score.
14. The method according to claim 12, comprising, in response to
the selection of a candidate word, the steps: updating the
precursor by appending the selected candidate word to the end of
the precursor and deleting the leading word; selecting a new list
of candidate words whose index matches the precursor; and prompting
the user to either select one of the displayed candidate words or
enter a desired word.
15. The method according to claim 12, comprising updating
periodically the score of selected said candidate words stored in
said dictionary to reflect a frequency of usage.
16. The method according to claim 12, wherein if no match between
at least one said index stored in said dictionary and said
precursor is found, then a new entry in said dictionary in which
said precursor is the index and the candidate word is the entered
word is created.
17. The method according to claim 12, wherein selected said
candidate words from the selected list whose initial character(s)
match the character(s) entered using a text entry device are
displayed.
18. Computer program stored on a computer-readable medium for
causing a processor to perform the method according to claim 12
when the computer program is executed by the processor.
Description
FIELD OF THE INVENTION
[0001] The present invention relates to a device incorporating an
improved text input mechanism, a method for improved text input and
a computer program for performing the method. More particularly,
the present invention relates to a device such as a personal
digital assistant or cellular telephone which incorporates a method
for contextual prediction of the next word of text to be input. The
method of the present invention minimizes the number of keystrokes
necessary to input text.
BACKGROUND OF THE INVENTION
[0002] The use of miniature computers such as personal digital
assistants (PDA) and cellular devices capable of text messaging has
become increasingly popular. Such miniature devices include
correspondingly miniature sized input mechanisms. Typically PDA's
and the like utilize a miniature keyboard or a multi-tap data entry
mechanism. The miniature size of the input mechanism makes it
impractical to enter large amounts of text, and it is desirable to
minimize the number of keystrokes necessary to enter text.
[0003] Prior approaches for minimizing the number of keystrokes
have focused on so-called disambiguating methods in which the
system refines its guess of the text being entered as the user
enters each letter of text. U.S. Pat. Nos. 5,818,437, 5,953,541 and
U.S. patent publication numbers 20020152203, 20020163504, and
20020126097 are representative of the state of the prior art.
[0004] The prior art approach is akin to thumbing through a
dictionary. The system "flips" to the appropriate section of the
dictionary once the initial letter is entered, and displays words
beginning with the initial letter. Due to limitations in the size
of the display, only the first n words are displayed. If the
desired word happens to occur within the first n words then the
user may select the desired word from the list of n words.
[0005] If the desired word does not appear, then the user enters
the second letter, whereupon the system displays words beginning
with the first two letters. If the desired word occurs within the
displayed list of n words then the user may select the desired word
from the list.
[0006] If the desired word does not appear, then the user then
enters the third letter, whereupon the system displays words
beginning with the first three letters. The user continues to enter
letters until either the desired word is displayed in the list of n
words or the word is fully entered.
[0007] The aforementioned approach has not sufficiently reduced the
number of keystrokes. Accordingly, what is needed is an improved
method for reducing the number of keystrokes necessary to enter
text.
SUMMARY OF THE INVENTION
[0008] Disclosed is a device incorporating a predictive text entry
mechanism, a predictive text input method and a computer program
for performing the method. The device includes a display screen for
displaying m-words of data, a text entry device for entering data,
a processor receiving data from the text entry device and causing
it to be displayed on the display screen, wherein upon activation
said processor initializes a precursor to a predefined value. The
device is provided with a non-volatile memory storing a dictionary
containing a plurality of entries, each such entry including an
index, a candidate word, and a score. The processor selecting a
list of n-number of candidate words from the dictionary whose index
matches the precursor (i.e. the context). The processor causing
m-number of candidate words from the list of candidate words to be
displayed on the display screen. The processor causing the display
to prompt the user to select one of the displayed candidate words
or enter a desired word using said text entry device.
[0009] In response to the selection of a candidate word, the
processor updates the precursor appending the selected candidate
word to the end of the precursor, deleting the leading word,
selecting a new list of candidate words whose index word matches
the precursor, and causing the display to prompt the user to either
select one of the displayed candidate words or enter a desired word
using the text entry device.
BRIEF DESCRIPTION OF THE DRAWINGS
[0010] FIG. 1 is a block diagram of a device including the
predictive data entry mechanism of the present invention;
[0011] FIG. 2 shows a sample device including the predictive data
entry mechanism of the present invention;
[0012] FIG. 3 shows one method of displaying candidate words;
[0013] FIG. 4 is a table showing a one-word dictionary according to
the present invention;
[0014] FIG. 5 is a table showing a two-word dictionary according to
the present invention;
[0015] FIG. 6 is a table showing normalization of the one-gram
dictionary of FIG. 4;
[0016] FIG. 7 is a table used to decode a word of text into a
grammatical representation; and
[0017] FIG. 8 is a table used to determine a candidate word using a
grammatical representation.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0018] The present invention provides an efficient method for
inputting text, and is particularly suited for cellular telephones,
handheld computers such as personal digital assistants, and other
devices requiring data entry.
[0019] The present invention is indifferent to the specific method
for inputting text. Text may be entered using a keyboard, a virtual
keyboard such as a touch screen, a so-called Graffiti style of
writing popularized by the PALM.RTM. operating system, or any other
man-machine interface.
[0020] The basic concept underlying the present invention is that
the next item of text is predicted based on the preceding word or
words of text. Furthermore, non-linguistic contexts (such as the
identification of a text field as "address field") may be used to
predict words.
[0021] According to a first embodiment, the system provides a list
of candidate words for the next word of text based on the preceding
word or delimiter. However, as will be explained below, the present
invention may readily be adapted to predict the next word based on
any combination of preceding words or delimiters and/or linguistic
analysis thereof.
[0022] FIG. 1 is a block diagram of the operating environment of
the present invention.
[0023] The system 10 of the present invention includes an input
device 100 which may be a keypad, keyboard, touch sensitive screen
or the like which hereinafter will be referred to as keyboard 100.
The system 10 further includes a display 102 for displaying the
text entered as well as an m word list of predicted words, the
screen may be an LCD screen or the like, and in the case of a touch
sensitive screen the display 102 may serve a dual purpose both as a
display and as the keyboard 100.
[0024] The present invention will work with any means for inputting
text, and any means for displaying text.
[0025] As further illustrated in FIG. 1, the system 10 includes a
processor or central processing unit (CPU) 104 which executes
commands via an operating system 108. The system 10 may include a
plurality of application programs 112, an application for decoding
handwriting into text 116, an application which displays text 114
and one or more dictionaries 110.
[0026] It should be appreciated that the text recognition
application 116, the display application 114, and one or more of
the dictionaries 110 may be implemented as firmware or microcode
stored on a non-volatile memory 120 such as an EPROM, EEPROM or the
like. Use of the term "software" throughout this disclosure should
be interpreted to include software implemented through firmware and
stored on a non-volatile memory medium.
[0027] The present system is realized as software or firmware or
the like, architecturally located between the operating system 108
of the computer 104 and the application 112 that receives text as
input, e.g. a word processor.
[0028] The system 10 is language specific and has one or more
dictionaries or vocabulary modules. The dictionary 110 and keyboard
100 are language specific. However, the software may allow for
switching dictionaries and thus switching languages.
[0029] A key aspect of the invention relates to its ability to
predict the next word the user will input based on the preceding
words or delimiters which will hereinafter be referred to as the
context or precursor. It is notable, that the context may also
arise from the type of application receiving the text or the type
of text field receiving the text. In the first case, the
application may give a hint to what kind of style the user will use
(SMS, short message service: colloquial writing vs. word processor:
formal writing). In the second case, some operating systems assign
unique identifiers to certain text fields such as "name",
"address", etc. This information may be used to activate a
different dictionary containing all names, addresses, cities
etc.
[0030] The system 10 predicts the next word the user will enter
based on the context, and displays a list of candidate words. The
number of candidate words displayed is dependent upon the size and
resolution of the display (the screen's real estate), and the font
size in which the text is displayed. Moreover, the system 10 may
optionally be configured to only display candidate words whose
score exceeds a threshold value, further limiting the number of
candidate words displayed.
[0031] FIG. 2 is a block diagram of an illustrative device 200
employing the system 10 of the present invention.
[0032] The device 200 includes display having a text edit field 202
showing a text string entered or selected by the user. The display
further includes a field 204 displaying a list of candidate words,
and optionally a field 206 displaying a virtual keyboard.
[0033] It is important to note, that the list 204 may take
different forms of presentation. For example, it may be realised as
a circle on which the words are presented. This is illustrated in
FIG. 3. This circle 204 may follow the position in the text where
the current input is directed to (this is called cursor which
usually takes the visual form of a blinking vertical bar). For the
sake of simplicity, the structure that is presented to the user
will be termed "list" in the following.
[0034] One of the important aspects of the invention is that the
list of candidate words is context dependent. The system 10 selects
candidate words by looking at the immediately preceding word or
words which will be termed a precursor. The number of words
defining a precursor may vary. It should be understood that the
term "word" as used throughout the specification refers to a
conventional word of text and to a delimiter such as a space, a
period, a comma, a semi-colon and the like.
[0035] As will be described below in further detail, the system may
iteratively reduce the size (measured in words) of the precursor in
order to obtain a desired number of candidate words. For example, a
three word precursor may have two corresponding candidate words. If
the display has room for 10 words, the system may obtain additional
candidate words by reducing the size of the precursor. A two word
precursor may for example have four corresponding candidate words,
and a one word precursor may have ten corresponding candidate
words.
[0036] The system 10 includes at least one dictionary 110 including
a list of candidate word entries; each candidate word entry
includes an index and a candidate word. It should be understood
that a given precursor may have many candidates. Thus, the same
index may appear in multiple candidate word entries in the
dictionary. For example the precursor "the" is likely to have many
candidates. Each candidate word entry further includes a score
indicative of the frequency of occurrence.
[0037] As will be explained below, the dictionary stored a
plurality of candidate word entries. Each candidate word entry
including an index, a candidate word, and a score. The size
(measured in words) of the index may vary from entry to entry. Thus
the dictionary may store candidate word entries having a one word
index, and entries having a two word index, and entries having a
three word index, etc. In this manner, the system may use a one
word precursor or a two word precursor etc.
[0038] Optionally, the score may be a normalized number between 0
and 1.
Bi-Gram Embodiment
[0039] The method and system of the present invention will now be
explained with reference to a bi-gram or two word embodiment in
which the precursor is the immediately preceding word (or
delimiter). In a tri-gram embodiment the precursor consists of the
two immediately preceding words (or word/delimiter combination).
Similarly, a four-gram embodiment the precursor consists of the
three immediately preceding words. It should be understood that the
teachings of the present invention may be implemented using any
number of words in the precursor. The size of the index stored in
the dictionary must correspond to the size of the precursor.
[0040] The dictionary 110 includes a plurality of candidate word
entries; each candidate word entry includes an index, and a
candidate word. The index is the key used to access the "candidate"
word(s). It should be appreciated that a single precursor may have
multiple candidate words. The index is at least one word or
delimiter in length, but may consist of two or more words.
[0041] According to the bi-gram embodiment each candidate word
entry in the dictionary 110 includes a list of two-word phrases in
which the first word is termed the index and the second word is
termed the "candidate" word. The precursor is the key or index used
to access the "candidate" word(s). It should be appreciated that a
single precursor may have multiple candidate words.
[0042] It should be appreciated that the precursor may be enlarged
to include any number of words or delimiters. For ease of
explanation the present example uses a one word precursor.
[0043] As noted above, each entry in dictionary 110 includes a
score indicative of the frequency of occurrence, i.e. the frequency
in which the user uses the phrase. A given precursor may have
multiple candidate words, and the score associated with each
candidate word may or may not be the same.
[0044] Each time the user enters (or selects) a precursor the
system 10 searches the dictionary 110 for corresponding candidate
words. The system 10 ranks the candidate words according to their
score, and displays the first m number of candidate words in field
204.
[0045] FIG. 5 below contains an illustrative list of two word
phrases.
[0046] The system 10 may treat the initial state before any word
has been entered in the same way it reacts when the precursor is a
delimiter such as a period. In this manner the system 10 may be
configured to provide candidate words in the initial state before
the user has entered a first character of the text.
[0047] The system 10 may be self-training in the sense that new
phrases and their corresponding scores may be dynamically added to
the dictionary 110. Moreover, the score of existing phrases may
dynamically update to reflect changes in the frequency of
usage.
[0048] According to a preferred embodiment, the system 10 is
provided with a robust dictionary 110 such that the full benefit of
the system is immediately available to the user. Moreover, if the
system 10 permits the score of existing phrases to dynamically
update then the dictionary 110 will self-optimize to reflect the
style, diction, and vocabulary of the user. Thus, even if the
dictionary 110 is initially empty, the system 10 will compile its
own dictionary reflective of the user's vocabulary, diction and
style.
[0049] An initial dictionary 110 may be compiled by examining a
body of text for x-word phrases, and ranking the frequency of the
x-word phrases. In the bi-gram embodiment of the present example x
will equal to 2.
[0050] Ideally the body of text used to compile the dictionary will
contain a rich vocabulary and varied examples of style and
diction.
[0051] The system 10 is especially useful for languages for which
it is complex to input words, e.g. Chinese. The system 10 may thus
operate on full symbols or on morphemes (one or more morphemes make
up a word of a language) or even on syllables (e.g. Hiragana or
Katakana in Japanese).
[0052] In the preceding example, the precursor was the immediately
preceding word (or delimiter). However, one of ordinary skill in
the art will appreciate that the invention may be readily adapted
to select one or more candidate words using any size of precursor,
e.g. a two or three word precursor.
[0053] In the aforementioned example, the candidate words were
generated using the score associated with each dictionary entry,
where the score is indicative of the frequency and is statistically
generated.
[0054] However, the candidate list can also be generated using
grammatical rules. For example, if the system 10 has identified a
context in which only a verb can follow, all nouns will be excluded
from the candidate list. If for example, one of the dictionaries
(c.f. to FIG. 7 as an example) is extended to have word classes
such as DETERMINER, ADJECTIVE, VERB, NOUN, etc. (these words are
capitalised to reflect the word class and not the word itself)
associated with words, then the precursor may be defined as a
string of word classes. For example, one precursor may be
"DETERMINER ADJECTIVE" and have "NOUN" as candidate. In this way,
the phrases "the red car" and "a little girl" may be realized
within one entry. The table in FIG. 7 is an example for a
dictionary that maps words into their respective categories. The
corresponding dictionary used to predict candidates from the
preceding context is given in FIG. 8. FIG. 8 is also an example of
varying sizes of context used for determining the precursor. While
there is a rule that predicts a NOUN after a DETERMINER, there is
also a rule that predicts a NOUN after a DETERMINER followed by an
ADJECTIVE. This rule would predict "car" and "girl" after "a
little". To determine a precursor, even more complex analyses may
be performed rather than a mere mapping of word forms to word
classes as is described below.
[0055] FIG. 7 provides a mapping mechanism for translating words
into their respective classes. Such a translation table is a very
simple mechanism of computing word classes from word forms. Other
grammatical predictions based on more complex linguistic analyses
of the context (rather than just mapping words to word classes) may
be readily adapted to generate the candidate list. It is
noteworthy, that the grammatical information need not be
constrained to the word class. Additionally inflectional
information maybe used (such as number: plural/singular, case:
nominative/accusative, etc.) to further specify the grammatical
information of the words. In this way NOUN.sub.SING, NOM (denoting
a noun in singular, nominative) may specify the word "girl", but
not the word "girls" (as this word is the plural form and would
correspond to NOUN.sub.PLUR, NOM). Rule based prediction for this
precursor having the rule "precursor: NOUN.sub.SING, NOM,
candidate: VERB.sub.SING, NOM", where VERB.sub.SING, NOM denotes a
verb in singular, nominative, would prevent the candidate "wants"
for the precursor "girls", but would allow the verb for the
precursor "girl".
[0056] Also the translation of a grammatical precursor into a
grammatical candidate class may be realized in more complex and
algorithmic ways than the simple table that is given in FIG. 8. For
example, a set of rules maybe iteratively or recursively applied
before using the grammatical representation of the text just as a
precursor.
[0057] A simple grammatical prediction system would generate the
candidate list in the following way: first, translate the text
written already into a grammatical representation (for example
using table 4 given in FIG. 7). Next match the last n grammatical
representations against the precursors in the table 5 (FIG. 8) and
determine the candidate. Finally, translate the candidate back to
word forms using table 4 (in FIG. 7).
[0058] It is noteworthy that several rules could match the
grammatical context. In this case, the list of candidates may be
constructed by combining the candidates resulting from the various
matches.
[0059] In this version of predicting the text, the prediction was
made using grammatical information (FIG. 7) and grammatical rules
(FIG. 8) instead of statistical information. Of course, these two
methods may be combined in order to obtain even better prediction.
For example, the words in table 4 may have additional information
about the frequency of use. In this way, a ranking of the
candidates within one grammatical category may be obtained.
[0060] Generally speaking, the invention provides two methods for
inputting a word of text. First, the user can type in words,
character-by-character, using the keyboard. Second, the user may
select a displayed candidate word. It should be understood that
reference to typing in a word includes any method of entering
text.
[0061] As noted above, the system may provide one or more candidate
words in the initial state before the user has provided a
precursor.
[0062] According to a further refinement of the present embodiment,
the system may refresh or update the candidate list in response to
character(s) entered by the user. More particularly, the system
responds to the user's entry of the first letter of the desired
word by displaying only those candidate words whose initial letter
corresponds to the entered letter. In this case the candidate words
are selected in accordance with the precursor and the characters
entered by the user.
[0063] It is important to appreciate that the aforementioned
refinement is significantly different from the prior art
disambiguation methods described in the background of the
invention. The prior art methods simply lookup words in a
conventional alphabetic dictionary whose initial letter(s)
correspond to those entered by the user. Since the conventional
dictionary is not context driven, the system will display words
which do not fit the context. In fact the system will respond with
the same list of words each time the user enters a given letter,
regardless of the context.
[0064] In contrast, the present invention selects candidate words
based on the context (precursor). The list of candidate words are
ordered according to the score and only the first m words are
displayed. The aforementioned refinement narrows the list of
candidate words by selecting only those candidate words whose
initial letter(s) correspond to those entered by the user. Thus the
refinement makes the score a secondary sort key rather than the
primary key. Most importantly, the present system will provide
suggestions for words following the word just entered. Thus, there
will be candidates for the next word although no key stroke has
occurred. None of the prior art has claimed such an ability. Prior
art only assists the user complete the current word. The present
invention predicts the word following the current word.
[0065] In the simplest case, the score mentioned above may be the
frequency of use.
[0066] Optionally, the processor will dynamically adjust the size
of the precursor to ensure that the system obtains a desired number
of candidate words. Thus, if the system retrieves too few candidate
words it can dynamically reduce the size (in words) of the
precursor. Alternatively, if the processor retrieves too many
candidate words it can dynamically increase the size (in words) of
the precursor. In other words, if the number of candidate words
retrieved is a threshold number of words greater than the number of
words which may be displayed on the screen then the processor can
dynamically increase the size of the precursor. Using our example,
the processor may shift from a bi-gram which uses a one word
precursor to a tri-gram which uses a two word precursor, i.e. the
two words which immediately precede the word to be entered.
[0067] Optionally, the dictionary 110 may include a list of
individual words (candidate words without an associated precursor)
and their respective score in addition to the aforementioned x-word
phrases. These single word entries will be termed orphan words. As
will be explained below, these orphan words are used to supplement
the candidate words when the number of candidate words falls below
a pre-defined threshold number. For example, if there is only one
candidate word for a given precursor, the system 10 may display one
or more of the orphan words.
[0068] The embodiments and refinements discussed above were
explained with reference to a single dictionary 110. However, the
present invention may be implemented using multiple dictionaries
110 without departing from the scope of the invention. For example,
a first dictionary 110 may be used to store single word (one-gram)
entries and their respective scores, and a second dictionary 110
may be used to store two-word (bi-gram) phrases and their
respective scores. Alternatively, the system 10 may include a
fixed, pre-defined dictionary 110 and a user-defined dictionary
110. Thus the concept of the invention may be extended to include
any number of dictionaries. Also, as mentioned above, different
dictionaries may be used for different receiving applications or
different text fields.
[0069] As computing power is generally limited in portable devices
it may be useful to keep the dictionaries 110 in different tables.
Generally, a one-gram table/dictionary will be smaller than a
bi-gram dictionary and thus can be searched faster. However
implementation details do not change the scope of the invention and
one of ordinary skill in the art will appreciate that there are
different possibilities to implement the dictionaries and their
access. For example the mapping from precursor to candidates may
not be realised through an index, but may arise from a more complex
computation. This may be the case for predictions based on
grammar.
[0070] One method for calculating the score for a dictionary will
now be explained with reference to the following example. Consider
the following text:
[0071] "The cat sat on the mat. The dog came along. The dog chased
the cat."
[0072] Dictionary Build Up:
[0073] Generally, the text is pre-processed, to include a leading
delimiter such as a period as the first character. Moreover, a
space is inserted between words and delimiters (periods, commas and
the like) in order to isolate the delimiters.
[0074] Afterwards, the above sentence would read:
[0075] ". The cat sat on the mat . The dog came along . The dog
chased the cat."
[0076] For building up a one-word (one-gram) dictionary, the words
and their frequency are counted. In the present example, different
word cases (differences in capitalization) are treated as different
words, i.e. "The" and "the" would be to different words.
[0077] Table 1 (FIG. 4) shows the one-word dictionary for the
example text.
[0078] For building up a two-word (bi-gram) dictionary,
combinations of two words and/or delimiters are used as entries for
the dictionary (c.f. Table 2, FIG. 5).
[0079] It is apparent from FIG. 5 (Table 2) that for bi-gram
frequencies, the text corpus used for training must be larger to
get better estimates of the frequencies. Here, most frequencies
have the value of one. Correspondingly for a tri-gram frequency in
which the precursor is two words the text corpus used for training
must be even larger still to get better estimates of the
frequencies.
[0080] Initial Word Candidates:
[0081] As described above, the system 10 may be configured to
present a list of candidate words in the initial state before the
user has entered a precursor. In the preferred embodiment these
candidate words are generated by looking at bi-grams, where the
precursor is in fact a period. Using the bi-gram dictionary above,
the only candidate: is "The", as there is only one bi-gram entry in
the dictionary starting with the period, namely ". The".
[0082] Obviously, it is desirable to have a full complement of
candidate words in order to provide a reasonable number of
selections from which to choose. According to one embodiment, any
empty slots in the candidate list may be filled by orphan
(one-gram) words in decreasing order of frequency. Words that
already appeared due to bi-gram frequencies will be excluded from
the fill-up items.
[0083] In an initial state which assumes a precursor ".", the
following eight item candidate list may be selected using the
one-gram and bi-grams (FIGS. 4 and 5): "The . cat the dog sat on
mat" The first candidate "The" is selected from Table 2 using the
precursor ".". However, since FIG. 5 does not contain any other
candidates for the precursor ".", the remaining seven words are
selected from the orphan word dictionary (one-gram) FIG. 4 (Table
1) according to their score.
[0084] Entering Text Using Candidates Only:
[0085] Continuing with the previous example let us assume that the
user selects the candidate word "The" and this word is then
displayed in the text output field 202. The system 10 now makes
"The" the precursor and selects a new list of candidates. The two
obvious candidates from the bi-gram dictionary given in table 2
would be "cat" and "dog". As "The dog" has the frequency 2, it will
rank higher after sorting. Again, the rest of the list would be
filled up by orphan (one-gram) words.
[0086] The new list presented in field 204 to the user would thus
read like: "dog cat . The the sat on mat". The user could now
choose the word "sat". The only item in the bi-gram dictionary
would then be "on", and so on.
[0087] Entering Text Using Candidates and Keyboard:
[0088] According to the aforementioned hypothetical, on start-up,
the system 10 generates the following initial candidate: "The . cat
the dog sat on mat". If the user enters a character instead of
selecting a candidate word, for example the character "t", then the
system filters out all words from the candidate list that do not
begin with the letter "t" independent of the capitalization of the
candidate words. In the present situation, there would only remain
two words that make up the reduced list: "The the". Again, the
residual positions would be filled up by orphan (one-gram) words,
in the present case (i.e. the present dictionaries in Tables 1 and
2) however, there are none.
[0089] Constructing A User Dictionary:
[0090] According to one refinement, the system 10 adapts to the
user's style by tracking the words that the user enters by building
up X-gram and one-gram frequencies, where X is any integer value
greater than 1. The X-gram and one-gram frequencies may all be
stored in a single dictionary 110. However, according to a
preferred embodiment, one dictionary 110 contains pre-defined
X-gram and one-gram frequencies and another dictionary 110 stores
user-defined X-gram and one-gram frequencies. The user-defined
X-gram and one-gram frequencies will be different in their
distribution from the pre-defined dictionary(s) 110 which were
computed from a different corpus of text. The text corpus may, for
example, include newspapers, stories, etc. and will thus differ
from the user's style. The text corpus used to construct the
pre-defined X-gram and one-gram frequencies may be very large
compared to the user's inputs. Consequently the frequencies of the
respective dictionaries 110 may need to be normalized to the
overall size of the corpora and combined to have a reflection of
the user's writing style in the candidate list.
[0091] Normalization of the one-gram dictionary (Table 1, FIG. 4)
will result in Table 3 (FIG. 6). Please note that the summed
frequency which by definition equals the overall size of the corpus
of the words is 19.
[0092] Note that, due to a division by a constant factor, the order
of frequencies remains the same. Note also, that all values range
between 0 and 1. Such normalization would be computed for all
dictionaries. Thus, independent of their size and absolute
frequencies, they would contain words with frequencies ranging from
0 to 1. When computing the combined score by which candidates are
ranked, now there would be a 1:1 influence between two dictionaries
on the ranking.
[0093] In the present embodiment, the score is computed for a
candidate word in the by the normalized frequency of the delivered
dictionary plus the normalized frequency of the user dictionary
times a scaling factor. The scaling factor accounts for the
stronger influence one would want to give to the user dictionary.
However, different formulae may be used to compute the overall
scores used for ranking the candidate words.
[0094] The normalization and computation of the score is analogous
for the bi-gram dictionaries.
[0095] Also, more than two dictionaries may be used for computing
the score. The procedure would be analogous to the aforementioned
procedure.
[0096] Although examples of an electronic device utilizing an
efficient context-related data entry method and its use have been
described and illustrated in detail, it is to be understood that
the same is intended by way of illustration and example only and is
not to be taken by way of limitation. The invention has been
illustrated in the context of a cellular telephone and personal
digital assistant. However, the data entry system may be used in a
data entry application. Accordingly, variations in and
modifications to the system and its use will be apparent to those
of ordinary skill in the art, and the following claims are intended
to cover all such modifications and equivalents.
* * * * *